Ken Raeburn [Thu, 3 Jun 2004 00:30:02 +0000 (00:30 +0000)]
The com_err library is now probably as thread-safe as it's going to get, on
UNIX; Windows still needs work....
* com_err.c: Include stdlib.h.
(com_err_hook_lock): New mutex.
(com_err_lock_hook_handle): New function.
(com_err_va, set_com_err_hook, reset_com_err_hook): Call com_err_finish_init,
and grab the lock.
* error_message.c: Don't include k5-thread.h.
(com_err_finish_init): New function.
(com_err_initialize): Initialize the new mutex.
* error_table.h: Include k5-thread.h.
(com_err_hook_lock, com_err_finish_init): Declare.
Ken Raeburn [Wed, 2 Jun 2004 23:47:36 +0000 (23:47 +0000)]
* k5-thread.h: Error out at compile time if thread support is enabled but the
thread system isn't pthreads. (This is after the code disabling thread support
for non-pthreads systems, so it's a placeholder for now that should never get
invoked.)
Ken Raeburn [Wed, 2 Jun 2004 23:41:51 +0000 (23:41 +0000)]
* prng.c (init_once): Variable deleted.
(krb5_c_random_add_entropy): Do the initialization once, using the yarrow_lock
mutex instead of k5_once to protect it.
Ken Raeburn [Wed, 2 Jun 2004 22:54:31 +0000 (22:54 +0000)]
* gss_libinit.c (gssint_lib_init): Initialize error tables here.
(gssint_initialize_library): Don't do it directly here. Make sure
gssint_lib_init has been called, and return the status.
(gssint_lib_fini): Remove error tables here.
(gssint_cleanup_library): Function deleted.
(initialized): Variable deleted.
Ken Raeburn [Wed, 2 Jun 2004 22:35:33 +0000 (22:35 +0000)]
Add/remove error tables in library initialization/termination, not per context.
Use only one library initialization scheme, and install the error tables in
only one place.
* krb5_libinit.c (initialized): Variable deleted, all references removed.
(krb5int_lib_init): Initialize error tables here.
(krb5int_initialize_library): Don't do it here.
* error_tables/Makefile.in (STLIBOBJS, OBJS, SRCS): Drop init_ets code.
* krb/init_ctx.c (init_common): Don't call krb5_init_ets.
(krb5_free_context): Don't call krb5_free_ets.
Ken Raeburn [Wed, 2 Jun 2004 22:25:37 +0000 (22:25 +0000)]
Move definitions of struct krb5_rc_st, struct _krb5_rc_ops, krb5_rc_ops, and
declarations of krb5_rc_register_type, krb5_rc_dfl_ops from k5-int.h to
rc-int.h. Include rc-int.h in the krb5/rcache files that need it.
Tom Yu [Tue, 1 Jun 2004 20:16:38 +0000 (20:16 +0000)]
fix buffer overflow in an_to_ln.c
* an_to_ln.c (rule_an_to_ln): Fix buffer overflow when parsing
principal names into components.
(do_replacement): likewise
(aname_replacer): Support error return from do_replacement
ticket: new
version_reported: 1.3.3
target_version: 1.3.4
tags: pullup
component: krb5-libs
Ken Raeburn [Sun, 30 May 2004 20:06:15 +0000 (20:06 +0000)]
* aclocal.m4 (KRB5_AC_GCC_ATTRS): On AIX 4, mark the destructor attribute as
always not working, until we can construct a good test for the order of
destructors in the multiple shared library case.
Ken Raeburn [Sun, 30 May 2004 07:27:32 +0000 (07:27 +0000)]
* getdate.y: Include krb5.h and kadmin.h.
(get_date): Drop second argument; always use NULL.
* kadmin.h: Include time.h and/or sys/time.h.
(get_date): Declare.
* kadmin.c (get_date): Don't declare.
(kadmin_parse_princ_args, kadmin_parse_policy_args): Drop second argument to
get_date.
* Makefile.in (LOCALINCLUDES): Define.
Ken Raeburn [Sun, 30 May 2004 06:05:54 +0000 (06:05 +0000)]
Silence some gcc compilation warnings:
* ksetpwd.c (verify_creds, get_init_creds_opt_init, userinitcontext,
init_creds): Now static.
(main): Remove one of two variables named "ccache".
Ken Raeburn [Thu, 27 May 2004 23:07:50 +0000 (23:07 +0000)]
* win-pre.in (CP): Copy nul: plus the file, to avoid carrying over the original
modification time, since some of our makefiles depend on $(CP) updating the
timestamp. Use binary mode for the copy, because default/ASCII seems to cause
a ^Z to be added to the file.
Jeffrey Altman [Wed, 26 May 2004 13:10:49 +0000 (13:10 +0000)]
* cc_mslsa.c: GetMSTGT(). Add krb5_context
parameter to allow krb5_get_permitted_enctype() to be called
instead of using a hardcoded list of enctypes which may change
in the future.
krb5_lcc_get_name(): fix return value if Kerberos is not supported.
Tom Yu [Tue, 25 May 2004 21:48:02 +0000 (21:48 +0000)]
* aesopt.h (PLATFORM_BYTE_ORDER): Treat _WIN32 as always
little-endian. Default to little-endian if there's no other
compile-time way to detect endianness, noting it as a guess.
(SAFE_IO): Error out if SAFE_IO is not set and endianness was
guessed.
Jeffrey Altman [Mon, 17 May 2004 14:48:14 +0000 (14:48 +0000)]
* kfw.nsi, kfw-fixed.nsi, utils.nsi:
- replace the UpdateDLL macro with the ReplaceDLL macro
- use ReplaceDLL to install all .exe and .dll files
this will allow the installer to work even when some files
are loaded by modules which cannot be terminated such as
Network Providers (afslogon.dll for example)
Jeffrey Altman [Sat, 15 May 2004 04:08:08 +0000 (04:08 +0000)]
2004-05-15 Jeffrey Altman <jaltman@mit.edu>
* cc_mslsa.c:
Do not use the FAILED() macro to test the result of ConstructTicketRequest().
ConstructTicketRequest() returns positive errors and FAILED() only considers
negative values to be a failure condition.
Also, close potential memory leak of LSA allocated memory.
Jeffrey Altman [Thu, 13 May 2004 03:16:24 +0000 (03:16 +0000)]
* send_tgs.c: krb5_send_tgs() was broken in the case of a KRB_ERROR
message. The krb5_response message_type field was never set
resulting in stack garbage being used instead. This would
break code which used transitive cross-realm to obtain service
tickets.
Ken Raeburn [Fri, 7 May 2004 23:29:38 +0000 (23:29 +0000)]
* configure.in: Check for endian.h and machine/endian.h.
* aes/aesopt.h (PLATFORM_BYTE_ORDER): Check for _MIPSEB, _MIPSEL. If endian.h
or machine/endian.h is available, include it instead of sys/param.h. Don't
mess around with multibyte character constants.
Ken Raeburn [Fri, 7 May 2004 23:26:43 +0000 (23:26 +0000)]
* configure.in: Check for machine/endian.h too.
* include/db-int.h: Include machine/endian.h if available. Check for
__LITTLE_ENDIAN__ and __BIG_ENDIAN__, _MIPSEB and _MIPSEL.
Ken Raeburn [Thu, 6 May 2004 02:28:25 +0000 (02:28 +0000)]
Start using our first bit of per-thread storage
* error_message.c (buffer): Static variable deleted.
(com_err_initialize): Register cleanup support for com_err thread-specific data
key.
(error_message): Use a per-thread dynamically-allocated buffer instead of
static storage, for the case where an unknown error code is given. If any
errors occur allocating or tracking the buffer, return a fixed message.
* t_com_err.c: Include stdlib.h. If TEST_THREADS is defined, include
pthread.h.
(run): Renamed from main, changed signature.
(main): New function. Just call run, or if TEST_THREADS is defined, create a
thread to call it.
Ken Raeburn [Thu, 6 May 2004 01:33:56 +0000 (01:33 +0000)]
Since the AES code builds, and doesn't do any configure-time byte order checks
that I noticed, something similar ought to work for the DB code. This is the
first cut; nightly testing builds should tell us if it's sufficient on most of
the platforms we work on.
* include/db-int.h: Include stdlib.h, and endian.h if available.
(LITTLE_ENDIAN, BIG_ENDIAN, BYTE_ORDER): If not defined, and if versions with
one or two leading underscores are defined, define the no-underscore form in
terms of the with-underscore one.
(DB_BYTE_ORDER): Define by checking LITTLE_ENDIAN, BIG_ENDIAN, and BYTE_ORDER;
report an error if that doesn't work. Don't check WORDS_BIGENDIAN.
* Makefile.in (all-prerecurse): Make sure headers generated by config.status
are up to date.
(include/config.h, $(srcdir)/include/config.h.in, include/db-config.h): New
rules.
* configure.in: Don't check byte order here. Check for endian.h.
Ken Raeburn [Wed, 5 May 2004 01:19:43 +0000 (01:19 +0000)]
* configure.in: Enable dependency support for shared lib.
* Makefile.in (test_et, t_com_err): Depend on, and link against, the new
support library.
(SHLIB_EXPDEPS, SHLIB_EXPLIBS, SHLIB_RDIRS, SHLIB_DIRS): New variables.
Ken Raeburn [Tue, 4 May 2004 19:38:18 +0000 (19:38 +0000)]
Make com_err tests exercise the shared-library support
* configure.in: Invoke KRB5_BUILD_PROGRAM and KRB5_RUN_FLAGS.
* Makefile.in (PROG_RPATH, PROG_LIBPATH, RUN_SETUP): New variables.
(test_et, t_com_err): Use the built library, even if shared.
(check-unix): Set up the environment properly to load shared libraries when
running the test programs.
Ken Raeburn [Tue, 4 May 2004 19:22:53 +0000 (19:22 +0000)]
* shlib.conf (alpha*-dec-osf*): Do remember to set use_linker_init_option and
use_linker_fini_option.
(mips-sgi-irix*): Set LDCOMBINE_TAIL only if not using gcc.
Ken Raeburn [Sat, 1 May 2004 01:43:15 +0000 (01:43 +0000)]
Use the new mutex in the replay cache structure
Rather coarse at the moment: While one thread is blocking waiting for data to
be flushed to disk, another thread is unnecessarily prevented from scanning the
in-memory data.
* rc_base.c (krb5_rc_resolve_type): Initialize the mutex in the replay cache
structure.
(krb5_rc_default, krb5_rc_resolve_full): Destroy it if creation of the replay
cache fails.
* rc_dfl.c (krb5_rc_dfl_get_span, krb5_rc_dfl_init): Lock the mutex while
operating on the replay cache object.
(krb5_rc_dfl_expunge_locked): Renamed from krb5_rc_dfl_expunge and made static.
Call krb5_rc_dfl_recover_locked.
(krb5_rc_dfl_expunge): New wrapper function, locks the mutex.
(krb5_rc_dfl_recover_locked): Renamed from krb5_rc_dfl_recover and made static.
Call krb5_rc_dfl_expunge_locked.
(krb5_rc_dfl_recover): New wrapper function, locks the mutex.
(krb5_rc_dfl_store): Lock the mutex. Call _expunge_locked.
Ken Raeburn [Sat, 1 May 2004 01:29:44 +0000 (01:29 +0000)]
Update mutex debug code to be somewhat compatible with multi-threaded
execution. It won't be as useful in single-threaded programs for detecting
bugs in the mutex handling for now, though.
* k5-thread.h (k5_mutex_debug_check_init, k5_mutex_debug_update_loc): New
macros.
(k5_mutex_debug_lock, k5_mutex_debug_unlock): Use them.
(k5_mutex_lock, k5_mutex_unlock) [ENABLE_THREADS && DEBUG_THREADS]: Use them
instead of k5_mutex_debug_lock and k5_mutex_debug_unlock.
(enum k5_mutex_debug_states): New enum.
(K5_MUTEX_DEBUG_LOCKED, K5_MUTEX_DEBUG_UNLOCKED): Change to enumerator values.
(k5_mutex_debug_info): Use the enum type. Reorder fields.
(K5_MUTEX_DEBUG_INITIALIZER): Update for new field order.
Ken Raeburn [Sat, 1 May 2004 00:00:02 +0000 (00:00 +0000)]
Add a mutex for the shared static storage used by Yarrow
* crypto_libinit.c (initialized): Variable deleted.
(cryptoint_initialize_library, cryptoint_cleanup_library): Use new macros for
automatic shared library init/fini functions.
(prng_cleanup): Declaration deleted.
(krb5int_prng_init, krb5int_prng_cleanup): Declare.
(cryptoint_initialize_library): Call krb5int_prng_init.
(krb5int_crypto_init): New function, checks that cryptoint_initialize_library
was called successfully.
(cryptoint_cleanup_library): Call krb5int_prng_cleanup only if the initializer
ran.
* crypto_libinit.h: Deleted.
* prng.c: Include k5-thread.h.
(init_once, init_error, yarrow_lock): New variables.
(krb5int_prng_init): New function; finish initializing the mutex lock.
(do_yarrow_init): New function.
(krb5_c_random_add_entropy): Call it once only. Lock the mutex before
processing the input.
(krb5_c_random_make_octets): Lock the mutex before extracting random bytes.
Ken Raeburn [Fri, 30 Apr 2004 00:59:37 +0000 (00:59 +0000)]
* shlib.conf (alpha*-dec-osf*): Add support for export list, using a temporary
file, and init/fini functions.
(mips-sgi-irix*): Previous support should not be used with gcc.
projects / thirdparty / krb5.git / log
